This Article Abstract Full Text (PDF) Services Download to citation manager PubMed PubMed Citation

Variations in Service Use in the Program of All-Inclusive Care for the Elderly (PACE): Is More Better?

¹ Department of Community and Preventive Medicine, University of Rochester School of Medicine, New York.
² Department of Medicine, Center for Health Policy Research, University of California, Irvine.

Address correspondence to Helena Temkin-Greener, PhD, Department of Community and Preventive Medicine, University of Rochester School of Medicine, Box 644, 601 Elmwood Avenue, Rochester, New York 14642. E-mail: helena_temkin-greener{at}urmc.rochester.edu

	Abstract

Top Abstract Methods Results Discussion Appendix 1 References

 
Background. To date, there has been little empirical evidence about the relationship between service use and risk-adjusted functional outcomes among the frail, chronically ill elderly population. The Program of All-Inclusive Care for the Elderly (PACE) offers a unique model within which to investigate this relationship. We examine variation in the risk-adjusted utilization of acute, rehabilitative, and supportive services in PACE, and assess whether use of these services is associated with risk-adjusted functional outcomes.

Methods. The analytical sample included 42,252 records for 9853 individuals in 29 programs, over 3 years. Outcome was measured as change in functional status. Service use was assessed for hospital and nursing home admissions, day center attendance, therapy encounters, and personal home care. Mixed regression, generalized estimating equation (GEE) log-linear Poisson models and bootstrap procedures were used.

Results. We examined the marginal effect of the five services on functional status over time, having controlled for each program's risk-adjusted use of services and functional status of their enrollees. We observed a statistically significant association between hospital admissions and functional status. Sites using more hospital care had worse functional outcomes. We found no other significant relationship between functional change and service use. However, correlations between program-level measures showed that sites providing more day center care and more therapy had significantly fewer hospital admissions.

Conclusions. Findings suggest that programs with high hospital use may do well to re-examine and adjust the intensity of day center care. Greater focus on service provision in this setting may enhance care coordination and lead to reductions in hospitalizations, better outcomes, and cost savings.

Key Words: PACE • Frail elderly persons • Service use • Functional outcome

Victor Fuchs, John Wennberg, and others have suggested that the organizational structure of medical care, dominant in the fee-for-service Medicare, creates clinical and financial incentives that often result in unwarranted variations in the use of medical services (10,13). According to Wennberg, a reduction in unwarranted variations can be achieved by implementing a "reform in financing that would facilitate investment in under funded aspects of care in the non-acute sectors and in the resources needed for active chronic disease management" (13).

The Program of All-Inclusive Care for the Elderly (PACE) was created within the framework of similar financial reforms (14). PACE is a community-based managed care program for frail, older, chronically ill individuals whose significant functional impairments make them eligible for nursing home care. The goals of PACE are to improve the coordination of acute and chronic care services to improve patient outcomes, prevent or delay institutional placement, and produce savings for public payers, Medicare, and Medicaid. PACE organizations are responsible for their enrollees' health care needs, ranging from preventive and primary to acute and long-term care. To this end, programs receive capitated funding from Medicare and Medicaid and have the ability to use these resources creatively, customizing services to fit the needs of individuals often in ways that are not possible under the traditional Medicare and Medicaid reimbursement rules. The program has been extensively described in the literature (15–17).

The purpose of this study is to address two questions. First, given the many commonalities of PACE such as a frail and disabled enrollee population, a delivery model based on an interdisciplinary team, similar financial incentives, and a shared philosophy of care, how much variation in service use is there across these programs nationally? Second, do enrollees in programs providing more (or fewer) services have better (or worse) functional outcomes?

	METHODS

Top Abstract Methods Results Discussion Appendix 1 References

 
Sample
The study was based on 29 (of 32) PACE programs that were in operation prior to January 2000 and agreed to provide their administrative database for the project. Limiting the sample to programs that started prior to 2000 ensured that only mature programs, the practice styles of which were not easily confounded by start-up learning curves, were included.

The final analytical file consisted of longitudinal data with 42,252 assessment records for 9853 individuals. For each program we used the most recent 3-year period of data available (end of the time period varied by program with the last date being December 31, 2002). Assessment records correspond to health and functional status assessments that are periodically conducted for all participants. Federal PACE regulations require all participants to be assessed at enrollment and thereafter semiannually, and following an unexpected health event or at a participant's request (18).

With the exception of the activities of daily living (ADLs), the dependent variable, if values of some health assessment variables were missing on one assessment, we assumed no change from the prior assessment period and carried those values to the current assessment (2.7% of observations). Another type of missing data may have occurred when whole assessment records were missing. We evaluated this by examining the time between assessments. The mean elapsed time between assessments was 4 months (SD = 1.5), with 90% of periods being between 2.4 and 6.3 months. Assessments with time differences >12 months accounted for 0.4% of all records. This shows that sites perform the assessments fairly regularly, and we found no significant differences between sites with regard to missing assessments.

Data and Variable Definitions
We estimated five models measuring each program's propensity to provide the following services: number of hospitalizations; number of short-term (length of service 90 days) admissions to nursing homes; day center attendance days per month; therapy encounters per months; and days of personal home care per month. For each program, we estimated one program-level health outcome, change in functional status of the participants over time. We measured functional status using an index of ADLs (19). The ADLs included in the data set are bathing, dressing, grooming, toileting, transferring, walking, and feeding. The data set defines each ADL as independent, needing help or supervision, or totally dependent on human assistance. We used this classification to assign numeric scores ranging from 0 to 2 for each ADL. The scores were then summed. Finally, we examined the association between each of the five services and program-level functional status.

Information about participants was obtained from a patient level administrative database collected by the programs. The data set included information about enrollees' demographics, health status, physical and cognitive functions, medical conditions and treatments, service utilization, and dates of death or disenrollment.

To risk-adjust for the case mix of the enrollees at different programs we included a number of explanatory variables (see Table 1). These variables were chosen because, based on our prior work (20–22) and on the review of the literature, we expected them to be important risk-adjusters for functional status. Among these we included selected diagnoses; vision, hearing, and communication impairments, and receipt of acute (e.g., injections, inhalations) or chronic (e.g., daily oxygen, ostomy) nursing treatments; bowel and bladder incontinence, and behavioral problems (measured on a scale of 0–2, with 0 indicating no impairment and 2 indicating total impairment). Other risk-adjustors included instrumental activities of daily living (IADLs): meal preparation, shopping, housework, laundry, heavy chores, managing money, taking medications, and transportation. Because of the possibility of reciprocal causation, the sum of IADLs was lagged, that is, at each assessment we used the value of IADLs from a prior assessment. Cognitive status was assessed using the Short Portable Mental Health Status Questionnaire (23), with severe cognitive impairment being measured by eight or more errors. We also controlled for the participants' age and the length of exposure each has had to PACE-provided care.

In the second stage we examined the relationship between programs' ADL performance and their propensities to provide these services by estimating a model in which program performance was the dependent variable and service use propensities were the independent variables.

First, we divided the analytical sample into two groups randomly assigning enrollees to each. One part of the data was used to estimate a mixed regression model for functioning, with the sum of the ADLs at each health assessment being the dependent variable. To model longitudinal observations for each person, corresponding to their periodic health assessments, we estimated the linear growth model with time from enrollment to each assessment as a predictor. This variable was interacted with PACE-site indicators to produce program-specific slopes, and may be interpreted as risk-adjusted measure of the programs' performance. The time variable was centered on the participant's mean so that the program-specific slope represented strictly longitudinal change for a person in that program. We included random effects for the participant's slope and the intercept to allow person-specific trajectory due to differences between people not accounted by observable characteristics. We used robust standard errors to account for possible remaining correlations between observations for the same person.

We used the second part of the sample to measure the programs' risk-adjusted utilization of the five services. We estimated mixed linear regression models for the monthly number of day center days, therapy encounters, and personal home care days, measured between consecutive assessments, as dependent variables. The independent variables were measured at the beginning of the periods and were centered on grand means. These models included fixed program effects as site-specific intercepts and simple random effects for the participants. To model the number of hospital and nursing home admissions we used generalized estimating equations (GEE) log-linear Poisson models. The logarithm of the length of the period was included in the model as the offset variable so that the program-specific intercept represented the logarithm of rate of admissions per year in the program. The working exchangeable correlation matrix and robust standard errors accounted for correlation between observations for the same person.

Finally, we modeled the association between the program performance (risk-adjusted functional change) and the propensities to provide the five services using a mixed regression model described in detail in Appendix 1.

	RESULTS

Top Abstract Methods Results Discussion Appendix 1 References

 
Table 1 shows means and standard deviations (SD) for all the variables used to estimate measures of program functional status performance and service use. The results of the mixed regression model for functioning and the models measuring programs' service use are made available in Appendix 2.

View larger version (17K): [in this window] [in a new window]   Figure 1. Ratio of each program mean to the overall mean for the Program of All-Inclusive Care for the Elderly (PACE) and coefficients of variation (CV) for each service

To determine the relative contribution of patient characteristics and program effects we decomposed the variation in the change (slope) in ADL dependencies into components attributed to the patient risk factors, which were included in the analysis and to program effects. The reduction in variance due to patient risk factors was 0.071 and the reduction due to program effects was 0.086, suggesting that programs play an important role in ADL trajectories.

The program-level model also provides estimates of correlations between program-level measures (bivariate associations) (Table 4). The correlations range from 1 (perfect correlation) to –1 (perfect inverse correlation); 0 indicates no correlation. Programs that provide more day center care have significantly (p <.05) fewer hospital admissions (–0.408). Programs with higher risk-adjusted rates of therapy services also have significantly fewer hospital (–0.291) as well as nursing home admissions (–0.406). Because therapy is provided in the day care it is not surprising that these programs also have greater propensity to provide day center care (0.167). Programs that provide more personal home care also provide more therapy (0.153) and experience fewer nursing home admissions (–0.169).

	DISCUSSION

Top Abstract Methods Results Discussion Appendix 1 References

 
Although PACE programs share many common features such as the patient population, the delivery system, and the financial incentives of a managed care model, they nevertheless exhibit wide variations in program performance. On a risk-adjusted basis, we document wide variations in functional status across the 29 sites, ranging from a statistically insignificant decline of 0.4 ADL dependencies to a statistically significant increase of 1.2 ADLs per year (i.e., deterioration in health status). We observe substantial differences across programs in risk-adjusted use of selected health care services. The least variation appears to be in the provision of day center care. Therapy encounters and hospital admissions exhibit somewhat greater variation. The most variable are the provision of nursing home admissions and personal home services.

Greater program use of these services does not appear to be associated with better participant outcomes in functional status. In fact, PACE programs providing higher intensity of hospital care demonstrate worse functional outcomes for their enrollees.

These findings raise several questions. What is causing the differences in service utilization? Why do additional services not lead to better outcomes? What are the policy implications of these variations in service propensities on the PACE programs? We offer some reflections on these issues in the following sections.

Differences in Service Utilization
Variations in the use of services, particularly hospital care, are most often attributed to the supply of hospital beds and to local physician practice styles (5). Although all PACE programs share a model of care that stresses the importance of controlling hospital utilization in managing PACE financial risks (15,17,24), studies have shown that even in this program the supply of hospital beds in the host community matters (25). Several PACE organizations have developed program-related housing, which offers access to transitional care beds. Having the option of transitional care may allow programs to prevent some hospital as well as short-term nursing home admissions, thus reducing the propensity to use these services. Early in the PACE development a number of programs had actively provided primary care in the home with the goal of delaying or preventing institutional care (hospital or nursing home). Over time, most programs have opted to provide primary care mostly in the day center setting. Programs may wish to re-examine this change in practice patterns particularly for the participants who are at highest risk of being hospitalized.

Conformity with local clinical practice style is also likely to impact propensities to institutionalize (in hospital or nursing home) program enrollees. Clinicians do not easily adopt departures from local practice, particularly in absence of scientific standards. Conformity with local practice may also be reinforced by "patients' expectations, which are partially formed by the treatment received by neighbors and friends" (10). PACE programs are not immune to the influences of such patient and/or family preferences (26).

One may expect patient and family preferences to matter even more in influencing day center or personal home care propensities. The day center is a setting in which a wide variety of services including primary care, restorative and recreational therapies, social work, and personal care are provided. Personal care services are also provided in the participants' homes to augment and sometimes to substitute for day center care. Each program, although undoubtedly influenced by patient preferences and the availability of family support, directly controls the use of these services. PACE organizations may choose to be more or less generous with regard to day center care. For example, programs may decide to expand their reliance on day center if, for whatever reason, they need to reduce the intensity of home health services. Programs may choose to limit day center attendance if they are experiencing rapid census growth, but are not yet positioned to open additional day facilities and/or are unable to adequately staff them. Our findings show that programs that rely more on care provided in the day centers tend to experience fewer hospital admissions.

Service Use and Quality of Care
Our study suggests that more is not necessarily better. We document that more hospital care was associated with worse functional outcomes in the PACE population.

Other studies have shown that when ADL needs of frail, older individuals are not properly met, their risk of hospitalization significantly increases (27–29). It also has been shown that hospitalizations for an acute illness tend to precipitate loss of function in older patients despite treatment of the acute illness (30,31). Geriatric patients admitted to the hospital have been shown to experience either a loss of or a diminished performance in at least one ADL. This decline in functional status occurs as early as the second day of the hospital admission (32,33). This study does not allow us to identify which of these paths is responsible for the observed relationship between hospitalizations and functional outcomes. PACE programs provide a continuum of services with an objective to maintain functional status and to prevent hospitalizations. Although they manage these services across institutional and community-based settings, the locus for these services is in the day center. The day center is a focal point for the provision of preventive and restorative services aimed specifically at persons with functional impairments. Our results show that more intensive use of day center care is associated with greater use of restorative care. Although the types of services provided in the day center are not unique, having these services consolidated in one place and coordinated by an interdisciplinary team of providers is. The provision of day center services, combined with the frequency of the interactions between providers and participants, also permits close monitoring and prompt interventions that likely reduce hospitalization in this population (24). A recent study comparing hospitalization in PACE and in the Wisconsin Partnership Program, a variant of PACE that does not rely on the provision of services through the day center model, showed that PACE enrollees had significantly fewer hospital admissions and emergency room visits (34). PACE programs with higher risk-adjusted hospitalization rates may do well to re-examine and adjust the intensity with which day center care is provided. Greater focus on service provision and coordination in the day center setting may lead to reductions in excess hospital admissions.

Programmatic and Policy Implications
Our findings demonstrate the need for improvements in the organization and delivery of care across PACE programs. The lessons from PACE can also be applied to other managed care programs serving frail, older individuals. By re-examining how and where care is provided, and by improving the targeting of services and settings to specific needs of the participants, programs can reduce excess hospitalizations and realize both better outcomes and cost savings.

These findings also pose a public policy challenge. Each day, health plans spend large amounts of money, mostly public funds, for services that are expected to increase or at least maintain the quality of care for their members. When evidence-based measures do not support that more care produces better outcomes, continuing widespread variations in service use are difficult to justify.

	APPENDIX 1

Top Abstract Methods Results Discussion Appendix 1 References

 
Statistical Analysis.-- In modeling the association between the program performance (risk-adjusted functional change) and the propensities to provide the five services, we assumed that these features were normally distributed latent random variables with joint distribution described by means and a covariance matrix. The program fixed-effect estimates, developed in the first stage of the analyses, measured these latent program features with error; that is:

where _p is the estimated fixed effect of program p in the functional decline model; Y_p is the true program effect ; _p is the "measurement" error with known variance; _i,p, X_i,p and v_i,p are analogous variables for each of the five (i = 1 ... 5) service models.

The means: m_y, m_x1 – m_x5 and the covariance matrix

of the latent variables Y_p, X_1p – X_5p can be translated (35) into the regression equation that depicts the relationship between the latent program performance Y_p and the latent propensities X_1p – X_5p:

We estimated the mixed model (36) for the equations (1) above with Y, X₁ – X₅ as random effects and used their estimated covariance parameters (as in equation 2) to calculate the service use coefficients b₁ – b₅ (as in 3). We used the bootstrap sampling technique (37) with 500 iterations to obtain 95% confidence intervals for these parameters. The correlation matrix for the program features was calculated based on the estimated variance–covariance matrix.

	Acknowledgments

Top Abstract Methods Results Discussion Appendix 1 References

 
This study was supported with funding from the National Institute on Aging Grant R01-AG17555.

We wish to thank the PACE programs for their participation in this study.

	Footnotes

Top Abstract Methods Results Discussion Appendix 1 References

 
Decision Editor: Darryl Wieland, PhD, MPH

Received July 24, 2007

Accepted October 18, 2007

	References

Top Abstract Methods Results Discussion Appendix 1 References

 

1. Welch HG, Wennberg DE, Welch WP. The use of Medicare home health care services. N Engl J Med. 1996;335:324-329.[Abstract/Free Full Text]
2. Fuchs VR, McClellan M, Skinner J. Area differences in utilization of medical care and mortality among U.S. elderly. National Bureau of Economic Research; 2001. Report No.: Working Paper 8628.
3. Welch WP, Miller M, Welch HG, Fisher ES, Wennberg JE. Geographic variations in expenditures for physicians' services in the United States. N Engl J Med. 1993;328:621-627.[Abstract/Free Full Text]
4. Medicare Payment Advisory Commission. Variation and Innovation in Medicare. Report to Congress. Washington, DC: MedPac; 2003.
5. Fisher ES, Wennberg DE, Stukel TA, Gottlieb DJ. Variations in the longitudinal efficiency of academic medical centers. Health Aff. 2004 October 7;hlthaff.
6. Fisher E, Wennberg J, Stukel T, Sharp S. Hospital readmission rates for cohorts of Medicare beneficiaries in Boston and New Haven. N Engl J Med. 1994;331:989-995.[Abstract/Free Full Text]
7. Ashton CM, Petersen NJ, Souchek J, et al. Geographic variations in utilization rates in veterans affairs hospitals and clinics. N Engl J Med. 1999;340:32-39.[Abstract/Free Full Text]
8. O'Connor GT, Quinton HB, Traven ND, et al. Geographic variation in the treatment of acute myocardial infarction: the Cooperative Cardiovascular Project. JAMA. 1999;281:627-633.[Abstract/Free Full Text]
9. Fisher ES, Wennberg JE, Stukel TA, et al. Associations among hospital capacity, utilization, and mortality of U.S. medicare beneficiaries, controlling for sociodemographic factors. HSR: Health Services Research. 2000;34:1351-1362.
10. Fuchs VR. Perspective: more variation in use of care, more flat-of-the-curve medicine. Health Aff. 2004 October 7;Web Exclusive(October 7, 2004).
11. Fisher ES, Wennberg DE, Stukel TA, Gottlieb DJ, Lucas FL, Pinder EL. The Implications of Regional Variations in Medicare Spending. Part 1: The Content, Quality, and Accessibility of Care. Annals of Internal Medicine 2003;138:273-287.[Abstract/Free Full Text]
12. Fisher ES, Wennberg DE, Stukel TA, Gottlieb DJ, Lucas FL, Pinder EL. The implications of regional variations in Medicare spending. Part 2: Health outcomes and satisfaction with care. Ann Intern Med. 2003;138:298.
13. Wennberg J., Variations in use of Medicare services among regions and selected academic medical centers Commonealth Fund Publication No. 874. http://www.commonwealthfund.org/usr_doc/874_wennberg_variation_medicaresvcs.pdf?section=4039, December 2005 (Last accessed March 27, 2008).
14. Bodenheimer T. Long-term care for frail elderly people—the On Lok model. N Engl J Med. 1999;341:1324-1328.[Free Full Text]
15. Mui A. The Program of All-inclusive Care for the Elderly (PACE): an innovative long-term care model in the United States. J Aging Social Policy. 2001;13:53-67.
16. Gross D, Temkin-Greener H, Kunitz S, Mukamel D. The growing pains of integrated health care for the elderly: lessons from the expansion of PACE. Milbank Q. 2004;82:257-282.[Medline]
17. Eng C, Pedulla J, Eleazer PG, McCann R, Fox N. Program of All-inclusive Care for the Elderly (PACE): an innovative model of integrated geriatric care and financing. J Am Geriatr Soc. 1997;45:223-232.[Medline]
18. Federal Register. PACE Provider Regulation. 42 CFR Part 460. 2002.
19. Katz S, Ford A, Moskowitz R, et al. Studies of illness in the aged: the index of ADL, a standardized measure of biological and psychosocial function. JAMA. 1963;185:914-919.[Abstract/Free Full Text]
20. Mukamel D, Peterson D, Bajorska A, et al. Variations in risk-adjusted outcomes in a managed acute/long-term care program for frail elderly individuals. Int J Qual Health Care. 2004;16:293-301.[Abstract/Free Full Text]
21. Temkin-Greener H, Bajorska A, Mukamel D. Disenrollment from an acute/long-term care managed care program (PACE). Med Care. 2006;44:31-38.[Medline]
22. Mukamel DB, Peterson D, Temkin-Greener H, et al. The Program of All-Inclusive Care for the Elderly (PACE): program characteristics and enrollees health outcomes. Milbank Q. 2007;85:499-531.[Medline]
23. Pfeiffer E. A short portable mental status questionnaire for the assessment of organic brain deficit in elderly patients. J Am Geriatr Soc. 1975;23:433-441.[Medline]
24. Chatterji P, Burstein NR, Kidder D, White AJ. Evaluation of the PACE demonstration: the impact of PACE on participant outcomes. Abt Associates, Inc. 1–61. 1998. Ref Type: Generic.
25. Temkin-Greener H, Mukamel DB. Predicting place of death in the Program of All-inclusive Care for the Elderly (PACE): participant vs program characteristics. J Am Geriatr Soc. 2002;50:125-135.[Medline]
26. Temkin-Greener H, Gross D, Mukamel DB. Advance care planning in a frail older population: patient versus program influences. Res Aging. 2005;27:659-691.[Abstract]
27. Covinsky K, Justice A, Rosenthal G, et al. Measuring prognosis and case mix in hospitalized elders: the importance of functional status. J Gen Intern Med. 1997;12:203-208.[Medline]
28. Fortinsky R, Covinsky K, Palmer R, et al. Effect of functional changes before and during hospitalization on nursing home admission of older adults. J Gerontol Med Sci. 1999;54A:M521-M526.[Abstract]
29. Sands L, Wang Y, McCabe G, et al. Rates of acute care admissions for frail older people living with met versus unmet activity of daily living needs. J Am Geriatr Soc. 2006;54:339-344.[Medline]
30. Covinsky K, Eng C, Lui L, Sands L, Yaffe K. The last 2 years of life: functional trajectories of frail older people. J Am Geriatr Soc. 2003;51:492-498.[Medline]
31. Wu A, Yasui Y, Alzola C, et al. Predicting functional status outcomes in hospitalized patients aged 80 years and older. J Am Geriatr Soc. 2000;48:S6-S15.[Medline]
32. Hirsch CH, Sommers L, Olsen A, et al. The natural history of functional morbidity in hospitalized older patients. J Am Geriatr Soc. 1990;38:1296-1303.[Medline]
33. Warshaw G, Sampson S, Matthias R, et al. Functional disability in the hospitalized elderly. JAMA. 1982;248:847-850.[Abstract/Free Full Text]
34. Kane R, Homyak P. Wisconsin Partnership Program Evaluation Focusing on Utilization, Cost and Quality of Care. Centers for Medicare and Medicaid Services; 2004. Report No.: Contract No. 500-96-0008, Task Order #3.
35. Johnson R, Wichern D. Applied Multivariate Statistical Analysis. 5th ed. Upper Saddle River, NJ: Prentice Hall; 2002.
36. Raudenbush SW, Bryk AS. Hierarchical Linear Models. Thousand Oaks, CA: Sage Publications; 2002.
37. Goldstein H. Multilevel Statistical Models. 3rd Ed. London: Arnold; 2003.

This Article Abstract Full Text (PDF) Services Download to citation manager PubMed PubMed Citation